4.5. Optical Selections of Distant Galaxies: "Photo-z's" and Ly Emission Lines

The case for the use of photometric redshifts - that is, redshifts based upon colors in 2 or (likely) more wave bands - has gradually strengthened since the mid-1990s. Most critically, we now expect fair precision from photometric redshifts and few catastrophic failures.

Stern & Spinrad (1999) compare spectroscopic and photometric redshifts in the HDF. The photometric redshifts are from the Stony Brook group (Fernandez-Soto, Lanzetta & Yahil 1999), and are determined by fitting the observed galaxy colors (long wavelengths only for really distant candidates) with redshifted spectral templates. These templates may be empirical, synthetic, or a hybrid. A second approach (Connolly et al. 1995) is purely empirical - having already a relationship between previously-observed galaxy redshifts and the observed total magnitudes (m) with color information (C) to boot. Then a derived redshift can be found from the multi-dimensional (m,C) pairs, used for training. More detail on these "template fits" can be found in the Stern and Spinrad review. Comparisons between photometric and spectroscopic determination in the HDF yield residuals typically around 0.1 for z at almost all redshifts.

Naturally the most important usage of such photometric redshifts is at very faint levels (m > 26.5). These numerous faint galaxies are well beyond the capabilities of 10-m class telescopes for spectroscopic redshifts. The danger here is that galaxies marginally detected in the red-optical I,z bands and perhaps also in J, H, K [1.2, 1.6, 2.2µm] can feign very large redshifts if their signal is just a noise incursion at I or z bands, slightly below the 1 µm observational limit of silicon-based CCDs. Since this topic is close to the kernel of this review, we note that Lanzetta et al. (1999) give some examples of faint, red photometric-z cases of difficult S/N. Their redshifts could exceed 6. Almost all of these ambiguous but potentially exciting cases have yet to be resolved. I speculate that better IR photometry (perhaps using the rejuvenated NICMOS camera on HST) would help in resolving that situation and perhaps suggest targets for future generations of near-IR spectrographs.

There is also a systematic problem at some level with color/redshift degeneracies; blue galaxies in general may show similar colors over a substantial intermediate z range. Prior information like the galaxy apparent magnitude can help decisively. This "Bayesian" procedure is illustrated by Benitez & Broadhurst (1999) for the HDF(N).

My personal recent experience with "I-drops" (implying a galaxy with only detectable flux at wavelengths above the I band, 8500 Å at the red edge) is that many of the spectroscopic candidates (15 to 20 targets per slitmassk) are very difficult due to their faintness (z ~ 25-26 mag) at longer wavelengths. A few also turn out to be low-luminosity galactic stars; these late M, L, and T class dwarfs turn up rather frequently. Since many of the candidates come from ground-based imaging, their image structure is not a very discriminating way to separate stars from QSOs from galaxies.

Most of the I-drops show a marginally detected red-color continuum, and thus add little to our initial appraisal. It turns out that approximately a quarter of the I-drops do eventually yield a redshift; about a third of these with the continuum discontinuity at Ly (₀ 1216 - the Ly "forest"). Two-thirds of the spectroscopically detected I-drop systems (with eventual redshifts) have a noticeable to strong Ly emission line. That usually yields an unambiguous redshift, as the reader can see with the illustrations in Weymann et al. (1998) and Fig. 4, here, by Spinrad, Stern, Dawson, Filippenko, and the GOODS team (z = 5.83).

Figure 4. A recent Keck spectrogram of a color-selected (I-drop) faint galaxy. The strong Ly emission line indicates a redshift z = 5.83. Also note the continuum discontinuity. The "spectral teams" were led by Spinrad and Filippenko, with reductions by Daniel Stern and Steve Dawson. This galaxy was originally selected by Mark Dickinson and the GOODS team.

As these pages were being written, two preprints crossed our desk. In the first, Lehnert & Bremer (2003) discovered 6 galaxies at 4.8 z 5.8. These galaxies were selected as photometric "R-drops" - that is, with little flux in the R-band and a flat spectrum at longer wavelengths. Follow-up spectroscopy with the VLT yielded accurate redshifts for these 6, with fairly strong Ly emission. Their largest redshift was z = 5.869 (see Table 3 in Section 1.4.7).

The second very timely contribution, by Kodaira et al. (2003) (a Subaru telescope team), used deep narrow-band near-IR images to locate potentially very distant Ly galaxies. The group also obtained a few spectra which lead to two fairly certain identifications. One line, with a symmetric line shape, is assumed to be Ly and in the other case it appears to be satisfactorily asymmetric, hence reliably Ly (see Section 1.4.6 for discussion of this point). The best spectrum is of SDFJ 132418.3 at z = 6.578. That would make this Ly galaxy the largest redshift of any individual system measured to date. The redshift is only slightly greater than that of HCM 6 A (z = 6.56) by Hu et al.(2002), and Hu, Cowie & McMahon (2002).

These very contemporary detections of galaxies beyond the "QSO-limit" of z = 6.4 show us that UV emission from galaxies is still present at the "tail" of the "dark ages". A future space-desideratum will be the galaxy morphology in the Ly line. We are interested in any extended neutral gas about the galaxy - via the scattered Ly emission from the central ionizing region (Haiman 2002 and references therein).

When the luminosity function of Ly emitters is extended to z ~ 6.5 (fainter galaxies have to be included) we should be able to extend the SFR density to that great distance. A sample of the near-constancy of the SFR density from z ~ 2 to z ~ 5 is illustrated in Fig 5 (from Iwata et al. 2003). The galaxies going into the computation of the SFR density are photometrically selected, using the top of the UV-luminosity function (M_UV -5 logh < -20). Interestingly, an attempt by D. Stern and the author to utilize serendipitously discovered Ly emitters at z ~ 5 yields a SFR density slightly higher than that of the z ~ 5 Iwata point in Fig. 5 (with considerable uncertainty). We view this as a possible coincidence, as these two methodologies may be sampling different populations. It is somewhat surprising that the relatively slight decline of the SFR density, noted by Iwata et al. (2003) should be maintained to z ~ 5. At that redshift the detected objects are effectively sub-galactic in size and probably rather modest in mass. At least temporarily, their M/L ratios must be quite low. Will that be true of most small sub-galactic systems?

Figure 5. Star-formation rate density as a function of redshift based on the UV-luminosity function with a magnitude limit MUV < -20. Triangles and diamonds are from Connolly et al. (1997) and Lilly et al. (1996), respectively. Squares represent data from Steidel et al. (1999) at <z> ~ 3 and and 4. The circle is the data of Iwata et al. (2003) at z = 5. Filled symbols indicate values without correction for dust extinction. Dust extinction was corrected following the prescription of Calzetti et al. (2000) with E(B - V) = 0.15 for all data points. Dust-corrected values are denoted by open symbols. Plot courtesy of I. Iwata.